Geoforum 93 (2018) 97–104
Contents lists available at ScienceDirect
Geoforum
journal homepage: www.elsevier.com/locate/geoforum
Rural-urban circularity in China: Analysis of longitudinal surveys in Anhui, T 1980–2009 ⁎ Chen Chena, , C. Cindy Fanb a Asian Demographic Research Institute, Shanghai University, 99 Shangda Road, Baoshan, Shanghai, China b Department of Geography, UCLA, Box 951424, Los Angeles, CA 90095, USA
ARTICLE INFO ABSTRACT
Keywords: Hundreds of millions of migrants from rural China circulate between their home villages and host cities. While Rural-urban migration existing research tends to focus on the question of permanent settlement in cities, the phenomenon of circularity China which has prevailed for decades is not well understood. For example, how often do migrants return, how long do Circular migration they stay before migrating again, and whether and how these behaviors have changed over time, are seldom Longitudinal study studied. Drawing from the longitudinal migration histories of 530 rural migrants from six villages in Anhui Gender Province and using multi-level Poisson regression models, this paper examines how rural-urban circularity has Generation changed since the 1980s. We found that migrants who first left for migrant work in the 2000s spent less time in the home location when they return, compared to those who first left in earlier decades. Male migrants return less frequently than female migrants; and younger migrants return less than older migrants. Migrants who have had high-school education, and who have young children, a spouse, and a high-quality house at the home location tend to return more frequently and spend more time when they return than other migrants. Women’s circularity is more sensitive to the number of dependent children and the decade of first out-migration than men; and men’s circularity is more sensitive to education level and generation than women. Our findings underscore circularity as a fundamental attribute of rural-urban migration in China, identifies the gender and generational differences in circularity, and highlights the social and household ties that sustain migrants’ motivation to re- turn/circulate.
1. Introduction contributed to changes in rural-urban circularity during this period? We use the longitudinal migration histories of 530 rural workers from six The hundreds of millions of rural Chinese who work in urban areas villages in Anhui Province to answer the above questions. By using two- are usually referred to as migrants but not circulators in spite of their level Poisson regression models, we analyze how the frequency and moving back and forth between their home villages and the cities (Fan, duration of return migration have changed over time and how they are 2016; Han et al., 2009; Schmidt-Kallert, 2009). While scholars have related to gender and individual and household characteristics. studied the prospect of migrant settlement in cities (Fan, 2011; Tang The next section reviews existing theories about circular migration and Feng, 2015; Zhu and Chen, 2010), research on circularity is limited. and highlights some of the factors that explain changes in circularity. It One reason is the tendency to focus on the question of permanent set- is followed by a brief overview of rural-urban migration and circularity tlement rather than circular migration and multilociaty (Fan, 2011; in China. Our empirical analysis focuses on villagers’ migration his- Schmidt-Kallert, 2009). For example, many studies highlight hukou as tories via descriptive statistics and modeling. The paper concludes with an impediment to migrants’ ability to stay in cities (Chan, 1996; Wu and a summary of our findings. Treiman, 2004). Another reason is the reliance on cross-sectional data that is less powerful than longitudinal data for analyses of circular 2. Research on circular migration migration. Informed by theories and research on circular migration and recent Circular migration is not new and has been widely practiced around studies on migration in China (see the next two sections), this paper the world, e.g., seasonal and cyclical migration in South Pacific Islands seeks to answer two main research questions: (1) How has rural-urban (Bedford, 1973, 1980; Hugo, 1982), Africa (Clark et al., 2007; Collinson circular migration changed since the 1980s? (2) What factors have et al., 2006; Potts, 2010; Elkan, 1967), Southeast Asia (Hugo, 1982,
⁎ Corresponding author. E-mail addresses: [email protected] (C. Chen), [email protected] (C.C. Fan). https://doi.org/10.1016/j.geoforum.2018.05.013 Received 11 July 2017; Received in revised form 17 March 2018; Accepted 10 May 2018 0016-7185/ © 2018 Published by Elsevier Ltd. C. Chen, C.C. Fan Geoforum 93 (2018) 97–104
1983; Goldstein, 1978, 1987), and India (Deshingkar and Farrington, older migrants spend more years outside Germany once they exit than 2009; Gidwani et al., 2003). However, conventional theories of human the middle-age migrants. Being married and owning a house in Ger- mobility tend to consider migration as a one-way move from an origin many reduce the frequency of circulation and the number of years to a destination rather than an interactive and iterative activity invol- away. It is important to note that factors which contribute to less fre- ving multiple sites and communities. Therefore, circular migration is quent circulation do not always lead to shorter time away from Ger- often understood as a temporary solution for low-skilled workers that many. For example, being female and better educated reduces the will eventually vanish with economic growth (Bedford, 1973; Hugo, number of exits but does not significantly affect the number of years 1982). away from Germany. Speaking the local language shortens the time At the same time, circular migration has increased between devel- away from Germany but does not affect the frequency of circulation. oping and developed economies (Constant and Zimmermann, 2012; Having a spouse living outside Germany encourages guest workers to Massey et al., 1994), within developed countries (DaVanzo, 1983; spend more years away from Germany once they exit the country, but it Gáková and Dijkstra, 2008), and among high-skilled workers (Hugo, does not significantly affect the frequency of circulation. While a 2008; Qin, 2015). The increased prevalence and magnitude of circular drawback of Constant and Zimmerman’s work is that it does not con- migration hints at its resilience and begs the question if such activity is sider whether circularity changes over time, its analysis of the fre- merely a transitional step toward permanent settlement. Theoretically, quency of circulation and the amount of time a migrant spends in dif- this question challenges the assumption that migrants are unidirec- ferent locations provides an appropriate analytical tool for this paper’s tional and their objective is always permanent settlement. Rather, mi- empirical analysis. grants may not intend to or may not be able to establish a new per- manent residence, and some may rather maintain a “multi-locational” 3. Rural-urban circularity in China livelihood for an extended period of time (Schmidt-Kallert, 2009). Four theories have informed research on circular migration. China’s level of urbanization was only 18% in 1978 and it sky- “Migration transition theory” and “dual labor market theory” focus on rocketed to 56% by 2015 (Jiang, 2013: 23; Wu, 2016), largely attri- the effect of economic structure. Migration transition theory considers butable to massive rural-urban migration. The “floating population,” circulation as a major form of migration in a “transitional society” referring to migrants who are living in places different from where they (Zelinsky, 1971). In this vein, circular migration occurs when small- are officially registered, amounted to about 300 million or about 22% holder agriculture, agricultural surplus labor, and urban-centered in- of the population in 2015 (NBS, 2015). Most of them are from rural dustrial development coexist, and when the demand for skilled labor origins working in urban areas, and circular migration is a prominent does not exceed the demand for unskilled labor (Guest, 1999). Trans- feature among them (Bai and He, 2003; Fan, 2011; Han et al., 2009). portation is an important structural factor: it improves as a society Specifically, these migrants do not tend to settle down in urban desti- develops, and such improvement helps facilitate circulation (Acevedo nations but instead circulate back and forth while maintaining a et al., 2004; Hugo, 1981). Dual labor market theory highlights capital household split between the city and the countryside (Fan, 2011, 2016). owners who use migrant labor in the secondary sector to adjust pro- Such practice has continued for years and even decades and has become duction according to economic cycles (Piore, 1980). Because there are a norm among rural Chinese. always uncertainties and fluctuations in economic activities, this theory It is widely accepted that Chinese migrants’ circularity is due to the predicts that circular migrants are always in demand in industrial so- household registration (hukou户口) system (Chan, 1996; Cheng, 2008; cieties. Li, 2003; Wang, 2013), which severely limits rural migrants’ access to The third theory, “social network theory,” emphasizes the perpe- social benefits such as subsidized housing, health care, and education in tuation of circular migration through networks of contacts that mi- host cities. Recent research has also highlighted circular migration as a grants have developed. These networks provide social support and re- household strategy that aims at maximizing household income while duce living costs in host areas, which helps return migrants to migrate guarding migrants’ economic and social resources in the countryside again (Fan and Stretton, 1984). This explanation is reinforced by the (Fan and Wang, 2008; Fan, 2009, 2011; Fan et al., 2011; Hu et al., theory of cumulative causation, which contends that migration sustains 2011; Zhu, 2003, 2007; Zhu and Chen, 2010). Most empirical studies on itself by fostering more migration (Massey, 1990). circular migration in China have focused on the presumed progression Lastly, the “new economics of labor migration theory” (NELM) from circularity to settlement, seeking to explain for a strong settlement highlights circular migration as a household strategy to maximize intention (Fan, 2011; Tang and Feng, 2015; Zhu, 2007; Zhu and Chen, household income and minimize risk, especially in developing econo- 2010) or a high probability of becoming a permanent resident (Hu mies where the capital market is under-developed, job opportunities are et al., 2011; Mendoza, 2008; Poston and Zhang, 2008; Sun and Fan, not sustainable, and the formal social security system is insufficient 2011). (Hugo, 1982; Stark and Bloom, 1985; Stark, 1991). Migrants move to However, due to the paucity of longitudinal data, we know little urban areas to earn higher incomes and send back remittances to sup- about the intensity of circular migration, whether and how it has port left-behind family members, who benefit from the low cost of changed over time, and what factors affect the frequency of circularity. living in rural areas. That they continue to reside and work in rural Longitudinal data is needed also because both migrants’ composition areas makes it easier for migrant family members to return if appro- and behavior have changed. First, new-generation migrant workers priate urban work is no longer available. The above theories point to (generally referring to those born after 1980) have accounted for more the importance of considering factors at all three levels to explain cir- than half of the migrant population since the early 2000s, and their cular migration: individual, household, and structural, as illustrated by proportion is growing (NBS, 2011; NPFPC, 2015). Second, unlike the research on international migration (Massey and Espinosa, 1997; 1980s and 1990s when migrants tended to leave the spouse behind, it is Massey and Pren, 2012; Bastia, 2011). In the next section, we shall increasingly common for migrants to pursue urban work together with describe the factors at all three levels in the Chinese context. the spouse (couple migration) and even bring their children along (fa- Methodologically and empirically, existing research has rarely ex- mily migration) (NBS, 2014; Zhou, 2004). Third, it appears that the amined the frequency and intensity of circularity. One exception is length of time that migrants spend in cities has increased over time Constant and Zimmerman’s work which documents and explains the (Duan et al., 2013; NPFPC, 2015). number of times a guest worker exits and the number of years he/she All the above changes seem to point to less frequent rural-urban lives away from Germany (Constant and Zimmermann, 2007). Using a circulation over time. Specifically, new-generation migrant workers 14-year panel dataset of guest workers, they found that the frequency of have expressed a stronger preference for urban life than older migrants circulation first decreases and then increases with age. Younger and (Tang and Feng, 2015; Wu and Xie, 2006); migrating with family
98 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104 members reduces the need to return to the home villages (Shang and the rural labor force in both counties; (3) Y and Z were representative of Yu, 2014); and migrants’ spending more time in cities implies increased Anhui counties that had sent out rural-urban migrants, in terms of stability there (Duan et al., 2013). Nevertheless, research has also economic development. In each county, three villages were selected to shown that new-generation migrant workers still tend to plan to represent different levels of economic development. Having said that, eventually return to the countryside (Yao, 2010; Yue et al., 2010), and all the villages selected share several similarities. First, per capita that migrating with family members does not significantly affect mi- farmland is small. Second, out-migration to work has been common- grants’ settlement intention (Fan, 2011). place and has become a normal way of life or even culture of the vil- A better understanding of rural-urban migration in China requires lagers. Third, most of the migrant workers from the villages have gone analyses that document and explain circular migration and its changes to the Yangtze Delta area, such as Suzhou, Wuxi, Ningbo, and Shanghai, over time. While research on circular migration tends to define it as for these destinations’ employment opportunities and relative proxi- repeated migration involving out-migration, return, and at least one mity. In each village, 25 households were selected: 15 households additional out-migration (Fargues, 2008; Hugo, 2005; Vadean and where at least one member had had migration experience by 1995; and Piracha, 2009; Wickramasekara, 2011), studies on China often treat 10 households where no one had had migration experience by 1995. In circular migration the same as temporary migration, meaning migra- other words, the survey included a total of 150 households from six tion without a change in hukou status (Connelly et al., 2010; Hu et al., villages in two counties in the province. Over time, these households 2011; Jia, 2006). In this paper, we define circular migration as having have become increasingly similar: by 2009 almost all had at least one at least one return (moving out from the home village for more than six family member who had had migration experience. months and returning for more than six months). This definition allows The survey includes information on members of the household, their us to include the new-generation rural workers in the analysis because demographic characteristics and migration histories. Notably, each in- they might not have had enough time to migrate again after their first terview includes also retrospective information, thus enabling a long- return, given their young age and recent onset of out-migration. itudinal record of both individuals and households. The narratives, As described earlier, factors at the individual, household and culminating into about 1000 pages of text, which after coding make it structural levels all affect migration and circularity (Bai and He, 2003; possible to document individuals’ migration history and other time- Ma, 2001; Massey, 1990; Zhao, 2002). In the context of China, in- varying variables for each year over their life cycles. While interviewees dividual characteristics including rural workers’ age, gender, genera- may not explain directly why they make certain mobility decisions, tion/cohort and education level have been shown to significantly affect such as why they decide to do migrant work and why they return at a their economic wellbeing as well as their settlement intention (Duan particular time, the longitudinal data permits statistical analyses that and Ma, 2011; Hu et al., 2011; Li, 1997; Xue, 2012). In particular, the assess the relationship between those decisions and factors at in- literature has highlighted gender as a key factor that shapes migration dividual, household and structural levels. behavior. For example, rural women’s likelihood to participate in mi- Our dataset includes all household members mentioned in the in- grant work declines sharply after getting married and especially after terviews, amounting to a total of 843 individuals. The centerpiece of having children, while men’s likelihood to do so is usually only con- the dataset is a continuous record of individuals’ migration status and strained by physical and age factors (Jacka, 2006; Yang and Guo, other characteristics in year-intervals from 1980 to 2009, when such 2000). Household characteristics including the number of children, information is available. Specifically, migration status is a dichotomous migration status of the spouse and children, availability of grand- variable coded as “inside” or “outside” for all individuals in the dataset. parents, housing and the size of farmland in the countryside, are im- This inside-outside binary largely represents also the dichotomy be- portant indicators of migrants’ considerations and strategies (Fan, tween rural and urban locations and between rural and urban work. 2011; Hagan et al., 2008; Wang and Zhao, 2013; Zhu, 2007). Structural The binary also is consistent with how rural Chinese typically speak in factors such as labor market demand and social and economic policies terms of “being outside that year” or “being inside (at home) that year.” which have changed over time also affect migration and circularity For the purpose of coding, when an individual stays within the (Chan, 2010; Chan et al., 2010; Wang, 2010). boundary of his/her home “town” (zhen or 镇), which may incorporate a number of villages, for more than six months continuously out of a calendar year, we code his/her migration status as “inside.” Even if an 4. Data and variables individual works in the town and not the home village, he/she very likely commutes to and live in the village and is therefore considered Research on circularity requires longitudinal information that most “inside.” Otherwise, the migration status is coded “outside.” By using census-type surveys do not provide. We are fortunate to have partici- the six-month criteria, we exclude short-term moves, such as returns for pated in and have access to a longitudinal study of 150 rural households the Chinese New Year. Being away for six months is also the standard in Anhui province, which was launched in 1995 by the Research Center definition of “migrant worker” used by China’s National Bureau of for Rural Economy (RCRE) of China’s Ministry of Agriculture. Semi- Statistics and by many migration studies on China (Knight et al., 2011; structured questionnaires guided face-to-face interviews with each of NBS, 2016). the 150 selected households. Our partnership with Renmin University To examine the frequency of circularity, in this paper we focus on enabled further interviews with the same households ten years later, in rural-urban migrants, defined as individuals who have rural hukou, are 2005. Further interviews between 2009 and 2014 supplemented in- within the working ages of 15–64, are not in school, prison or the 1 formation since 2005 and up to 2009. Such repeated interviews al- military, and had moved outside at least once after age 15. Of the 843 lowed us to create a panel dataset for the period from 1980 up to 2009. individuals in the dataset, 530 are rural-urban migrants using this de- Anhui was chosen for our study because it was (and still is) a major finition. Following Constant and Zimmermann’s (2007) work on cir- sending province of migrant workers. When the project began in 1995, cular migration of immigrants in Germany, we compute the number of two counties, referred simply as Y and Z and located in respectively the returns a rural migrant has undertaken as well as the number of years northwest and southeast of Anhui, were selected based on three criteria: inside during an observation period. A return refers to the change of (1) compared with other counties, Y and Z had relatively long histories migration status from outside to inside. An observation period refers to of out-migration; (2) out-migration accounted for at least 20 percent of the period between the year when a subject turns 15 and the year of observation, which is always some year between 1980 and 2009. At the end of each observation period, we record the number of returns and 1 While the original survey included also Sichuan province, data on Sichuan is not included in this paper because the quantitative coding of those interviews is not yet the number of years inside a rural migrant has had during the period, as available. well as his/her individual characteristics and household characteristics.
99 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104
We refer to this record as an “observation.” The vast majority of rural documented that even after years of farmland conservation and ex- migrants in the dataset have more than one observations. pansion, the per capita cultivated land in Anhui was only about 1.30 The number of returns and number of years inside serve different mu,2 lower than the national average of 1.52 mu (Department of Land analytical purposes. The former documents the intensity of circularity. and Resources of Anhui Province, 2014). Second, most of the migrant When the number of returns changes from 0 to 1, a migrant completes a workers from the villages have gone to the Yangtze Delta area, such as circuit of migration. The more returns a migrant conducts within a Suzhou, Wuxi, Ningbo, and Shanghai, for these destinations’ employ- certain period, the more frequently the migrant circulates. The number ment opportunities and relative proximity. Third, most migrants work of years inside, on the other hand, represents the relative importance of in low-paid and low-skilled jobs such as construction, factory manual the hometown to rural-urban migrants. work, and domestic work. Nevertheless, if longitudinal data with suf- The 530 rural migrants contribute in total 5815 observations. The ficient information about such variables is available, the analysis would observation period ranges from one year to 30 years, averaging have been more complete. 11 years. Of all the observations, 3637 (62.6%) record no returns, re- presenting young migrants who have not yet returned or individuals of 5. Descriptive statistics any age who have never returned during the period recorded. Our ex- amination of the data shows that most are from the former group. Of all Table 1 presents the summary statistics (means and standard de- the observations, 1809 (31.1%) show one return; 320 (5.5%) two re- viations) of the selected variables. The statistics are tabulated sepa- turns; 39 (0.7%) three returns; 10 (0.2%) four returns. rately for the entire sample, male migrants, and female migrants. To examine how rural-urban circularity has changed over time, we Among the 5815 observations, 39% began the first out-migration in the fi “ fi ” have identi ed decade of the rst out-migration as an explanatory 1980s, 40.5% in the 1990s, and 20.5% in the 2000s. Male migrants variable. If the intensity of rural-urban circularity has indeed declined accounted for 64.1% of the observations. The average age of the ob- fi over time, then we would expect migrants who undertook their rst servations is 32.8, and among them 34% are of the old-generation, 49% migration in recent decades to return fewer times and spend shorter mid-generation, and 17% new-generation. The modal education level is time when they return than migrants in earlier decades. junior high which accounts for 52.4% of the observations. In terms of We also include a standard set of individual and household char- household characteristics, by the end of the observation period, mi- acteristics as independent variables. Individual characteristics include grants on average have 0.7 young children inside, 36% of them have a age, age squared, gender, generation, and education level. Age refers to spouse inside, 60.9% have a parent inside, 23.1% own a brick-and-earth the age in the year of observation. Based on the birth year, a rural house, 35.8% own a red brick house, and 41% own a brick-and-con- migrant is further categorized as old-generation (born before 1965), crete house. In terms of the home village, 16.8% of the observations are mid-generation (born between 1965 and 1979), or new-generation from Village Y1, 18.2% from Y2, and 21.3% from Y3, all in the northern (born in or after 1980). Education level records the highest education part of Anhui. From the southern part of the province, Village Z1 ac- level an individual has obtained by the year of observation. counts for 15.6%, Z2 13.2%, and Z3 14.9% of the observations. On fi Four household characteristics are included. The rst three re- average, each migrant has had 0.5 returns and been inside for about present the household arrangement in the year of observation: the two years. The average values of the two exposure variables—the number of dependent (younger than 15) children who are inside, maximum possible number of returns and the length of ob- whether the subject has a spouse inside and whether the subject has a servation—are 4.7 times and 9 years respectively. parent inside, the latter two being dummy variables. The fourth vari- The male and female samples of our data appear to be different in able is rural housing, which refers to whether, in the year of the ob- many aspects. The majority of the male migrants (53.0%) first out- ’ “ ” servation, the migrant s household has a brick-and-earth house migrated in the 1980s, while the majority of female migrants (52.1%) 砖土房 “ ” 红砖房 (zhuantu fang ); a red-brick house (hongzhuan fang ); or a first out-migrated in the 1990s. The 2000s accounted for 33.9% of fe- “ ” 砖混房 brick-and-concrete house (zhuanhun fang ). Brick-and-earth male migrants’ first out-migrations and only 13% of male migrants’ first houses are usually not in good condition and are considered the least out-migrations. Along the same vein, higher proportions of female desirable; red-brick houses are newer and more desirable; and brick- migrants than male migrants belong to the mid and new-generation and-concrete houses tend to be the newest and most preferred. Migrants groups, and male migrants are on average three years older than female “ in our survey frequently mentioned building a brick-and-concrete migrants. These differences show that in general rural women’s parti- ” house as an important reason for both migration (to increase income) cipation in out-migration occurred more recently than men’s. On ff and return (to build the house). To control for the potential e ect of the average male migrants are better educated than female migrants. fi home village, we include also ve village dummy variables. Female migrants have fewer young children inside, are less likely to It is important to note that the length of observation period is dif- have a spouse inside, and are much more likely to have parents inside ferent across observations in our sample. The longer an individual is than male migrants, all pointing to caregiving responsibilities that tend “ ” observed, the more years he/she is exposed to the risk of return and to deter women’s migration more so than men’s. Female migrants are ff being inside. Thus, the length of observation period must be o set. To less likely to have a red-brick house or brick-and-concrete house than ff that e ect, we include two exposure variables in the data estimation. male migrants, suggesting that the former are in greater economic need The length of observation period is used as the exposure variable for than the latter. On average, male migrants return 0.5 times and spend “ modeling the number of years inside. We also create the maximum 2.1 years inside, and female migrants return 0.4 times and spend ” possible number of returns (during the observation period) for each 1.7 years inside. Male migrants’ maximum possible number of returns is observation and use it as the exposure variable for modeling the 5.3 times, compared to 3.7 times for female migrants. These differences number of returns. may in part reflect the different lengths of observation by gender: male Because the primary data from the survey is in the form of narra- migrants are observed about 3 years longer than female migrants. tives and not from a formal questionnaire, it is limited by incomplete information for certain variables such as migration distance, migrant 6. Modeling circularity jobs and size of farmland, which can potentially affect frequency and duration of return. Having said that, all the six villages share simila- Since our panel data comprises observations of the same subject rities that suggest that the absence of these variables has not unduly over time, in order to predict the dependent variables, it is necessary to affected our results. First, per capita and per household farmland is small. Although no official data exists about the size of farmland in Y and Z, a report published by Anhui provincial government in 2014 2 1mu≈ 666.7 square meters.
100 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104
Table 1 Selected sample characteristics.
Variables All Male migrants Female migrants
Mean SD Mean SD Mean SD
Time Decade of the first out-migration 1980–1989 0.390 0.488 0.530 0.499 0.140 0.347 1990–1999 0.405 0.491 0.340 0.474 0.521 0.500 2000–2009 0.205 0.404 0.130 0.337 0.339 0.473
Individual Characteristics Male 0.641 0.480 Age 32.777 11.437 33.827 11.826 30.900 10.450 Age squared 1205.119 853.824 1284.072 901.544 1063.980 740.617 Generation Old-generation 0.340 0.474 0.410 0.492 0.214 0.410 Mid-generation 0.490 0.500 0.469 0.499 0.529 0.499 New-generation 0.170 0.375 0.121 0.326 0.257 0.437 Education level Below primary school 0.091 0.288 0.060 0.238 0.147 0.354 Primary school 0.278 0.448 0.232 0.422 0.360 0.480 Junior high 0.524 0.499 0.575 0.494 0.433 0.496 Senior high 0.106 0.308 0.132 0.339 0.060 0.237
Household Characteristics No. of children (age < 15) inside 0.669 0.911 0.717 0.936 0.582 0.858 Having a spouse inside 0.360 0.480 0.479 0.500 0.146 0.353 Having a parent inside 0.609 0.488 0.582 0.493 0.658 0.474 Rural housing Brick-and-earth 0.231 0.422 0.188 0.391 0.308 0.462 Red-brick 0.358 0.480 0.373 0.484 0.332 0.471 Brick-and-concrete 0.410 0.492 0.439 0.496 0.360 0.480
Village Y1 0.168 0.374 0.178 0.382 0.150 0.357 Y2 0.182 0.386 0.173 0.378 0.199 0.399 Y3 0.213 0.409 0.209 0.407 0.219 0.414 Z1 0.156 0.363 0.164 0.371 0.140 0.347 Z2 0.132 0.339 0.138 0.345 0.123 0.328 Z3 0.149 0.357 0.138 0.345 0.169 0.375
No. of returns 0.448 0.647 0.483 0.678 0.386 0.582 No. of years inside 1.965 3.875 2.122 4.113 1.686 3.389 Maximum possible no. of returns 4.740 3.327 5.304 3.529 3.732 2.648 Length of obs. 8.958 6.666 10.089 7.069 6.934 5.307
Sample size 5815 3729 2086 use a model that can accommodate time-varying variables. We chose 7. Modeling results two-level random-intercept Poisson regression models (Rabe-Hasketh and Skrondal, 2012) with four specific steps. First, our dependent Table 2 summarizes the regression results at the observation level, variables are counts, which are always non-negative integers. Since with the individual-level variance controlled. Models (1), (3), (5) and counts follow a Poisson distribution and our dependent variables show (7) use the number of returns as the dependent variable; and models no over-dispersion,3 we use Poisson regressions as our count model (2), (4), (6), and (8) use the number of years inside as the dependent (Gardner et al., 1995). Second, we adopt two-level models in recogni- variable. Models (1) to (4) were estimated using the entire sample, tion of the fact that different subjects may respond to various combi- models (5) and (6) the male sample, and models (7) and (8) the female nations of factors differently. The first level is the observation of a rural sample. All eight models are statistically significant. To conserve space migrant, which may account for some of the variance in the dependent and for the sake of interpretation, we show the regression results as variables. The second level is the rural migrant, whose characteristics incident rate ratios (IRR). may also account for some of the variance. Third, by using a random- Models 1 and 2 include only “decade of the first out-migration” as intercept, we allow the probability of return to vary across different the independent variable and village dummies as control variables. migrants. Finally, we tested for multicollinearity by running an or- These two models suggest that migrants who first migrated in recent dinary least squares regression using the same dependent and in- decades, especially after 2000, have undertaken fewer returns and dependent variables. The variance inflation factors suggest that corre- spent shorter time inside than those who first migrated in earlier dec- lations among independent variables have not unduly biased the ades. However, when other individual and household variables are models’ estimates. We estimated the models using STATA 14.0. included, in models 3 and 4, the decade effect is no longer significant except for migrants who first out-migrated after 2000 who spent sig- nificantly shorter time inside when they return compared to migrants who first out-migrated in the 1980s. Models 3 and 4 also suggest that holding everything else constant, factors which affect the number of returns and factors which affect the length of time a migrant spends inside are not always the same. Being male decreases the incident rate 3 We calculated the dispersion parameter (alpha) for each dependent variable. They of return by 51% and the percentage of years inside by 66%. Aging by have an alpha of 0, which indicates no over-dispersion.
101 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104
Table 2 Two-level Poisson regression results.
Entire sample Male migrants Female migrants
No. Returns No. Years inside No. Returns No. Years inside No. Returns No. Years inside No. Returns No. Years inside Model (1) Model (2) Model (3) Model (4) Model (5) Model (6) Model (7) Model (8)
Level 2 Variables Time Decade when the first out-migration started 1990–1999 0.820 0.488* 1.229 0.824 1.729 1.423 0.572 0.204** (0.230) (0.198) (0.352) (0.342) (0.628) (0.735) (0.279) (0.155) 2000–2009 0.335*** 0.0909*** 0.740 0.245*** 1.325 0.504 0.385* 0.0709*** (0.105) (0.0408) (0.264) (0.124) (0.738) (0.388) (0.200) (0.0568)
Individual Characteristics Male 0.490*** 0.339*** (0.121) (0.119) Generation Mid-generation 0.355*** 0.345*** 0.280*** 0.253** 0.652 0.686 (0.101) (0.138) (0.108) (0.136) (0.261) (0.397) New-generation 0.137*** 0.169*** 0.0432*** 0.0392*** 0.465 0.856 (0.0558) (0.0942) (0.0294) (0.0358) (0.234) (0.596) Education level Primary school 1.426 2.029 1.274 2.087 1.248 1.655 (0.580) (1.187) (0.922) (2.140) (0.532) (1.091) Junior high 1.232 1.674 1.356 2.189 0.894 1.083 (0.502) (0.981) (0.961) (2.196) (0.395) (0.734) Senior high 2.699** 5.771** 4.430* 13.82** 1.255 2.003 (1.332) (4.064) (3.447) (15.20) (0.810) (1.920)
Level 1 Variables Individual Characteristics Age 0.992 1.056*** 1.005 1.042*** 0.948 1.081*** (0.0162) (0.00924) (0.0191) (0.0106) (0.0322) (0.0215) Age squared 1.000 1.000** 1.000 1.000 1.001* 1.000 (0.000207) (0.000106) (0.000237) (0.000121) (0.000464) (0.000266)
Household Characteristics No. of children (age < 15) inside 1.075** 1.077*** 0.997 1.020 1.284*** 1.229*** (0.0326) (0.0158) (0.0374) (0.0189) (0.0694) (0.0322) Having spouse inside 1.446*** 1.341*** 1.501*** 1.395*** 1.479*** 1.295*** (0.0928) (0.0441) (0.120) (0.0602) (0.163) (0.0675) Having a parent inside 0.910 1.060 0.935 1.057 0.901 1.133 (0.0846) (0.0506) (0.102) (0.0615) (0.161) (0.0960) Rural housing Red-brick house 1.211 1.490 1.491 1.862 0.971 1.115 (0.324) (0.544) (0.580) (0.963) (0.331) (0.556) Brick-and-concrete house 1.650** 2.471*** 2.030** 3.113*** 1.501 2.214** (0.327) (0.562) (0.561) (0.952) (0.415) (0.757)
Village Y2 1.454 1.495 1.348 1.336 1.130 0.887 1.419 1.789 (0.564) (0.823) (0.510) (0.723) (0.596) (0.660) (0.723) (1.375) Y3 1.459 1.376 1.379 1.211 1.577 1.345 1.218 1.165 (0.545) (0.728) (0.493) (0.617) (0.787) (0.945) (0.577) (0.831) Z1 1.520 1.673 1.468 1.644 0.845 0.708 2.668* 4.532** (0.608) (0.948) (0.574) (0.915) (0.471) (0.555) (1.371) (3.492) Z2 0.573 0.440 0.518 0.413 0.257** 0.169** 1.141 1.155 (0.250) (0.269) (0.218) (0.247) (0.157) (0.144) (0.616) (0.936) Z3 1.143 1.152 1.187 1.354 0.818 0.741 1.570 2.351 (0.474) (0.678) (0.464) (0.755) (0.468) (0.596) (0.774) (1.751) Constant 0.0273*** 0.0273*** 0.0528*** 0.00496*** 0.0203*** 0.00213*** 0.114** 0.00387*** (0.00929) (0.0132) (0.0326) (0.00374) (0.0177) (0.00236) (0.107) (0.00430)
Model Chi-square 21.52 37.10 135.53 452.56 96.28 261.72 69.77 244.51 Number of observations 5815 5815 5815 5815 3729 3729 2086 2086 Number of migrants 530 530 530 530 283 283 247 247
St.d in parentheses. * p < 0.1. ** p < 0.05. *** p < 0.01. one year increases the incident rate of being inside by 5.6%, but does generation migrants, being a mid-generation migrant reduces both the not affect the frequency of returns. The relationship between age and incident rate of return and being inside by about 65%. Also compared to the incident rate of being inside is linear, reflected by the significant the old-generation, being a new-generation migrant reduces the in- coefficient of age-squared as 1.0. This suggests that as a migrant gets cident rate of return by 86% and being inside by 83%. As for education, older, he or she will spend more time inside. Compared to old- only the senior high level affects circularity. Compared to a migrant
102 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104 who has no formal education experience (below primary school), a Household characteristics such as having more dependent children in- migrant who has attended senior high has an incident rate of return that side, having a spouse inside, and having a high-quality rural house also is 2.7 times higher and an incident rate of being inside that is 5.8 times increase the frequency of return and the length of time inside. higher. This result may be related to the fact that better-educated mi- Comparing men’s and women’s circularity, we found that genera- grant workers are likely to make use of their migration experiences in tion and education level affect only men but not women, while decade their hometowns more easily than migrants with lower education level. of the first out-migration affects only women but not men. These results Our interviews show that among rural migrants with senior high edu- suggest that rural women’s participation in migrant labor force is more cation level who have returned, some have tried to run small businesses sensitive to increase in job opportunities in urban areas, while men’s in the hometown and others are employed in nearby towns as civil participation in migration work is more affected by their generational servants or village officers. Being closer to the family is apparently a characteristics and human capital. reason for their return, despite the fact that their income may suffer Gender division of labor persists, as indicated by men’s spending compared to migrant work. less time inside and women’s returning to the countryside more often, Models 3 and 4 show that in terms of household characteristics, probably related to giving birth and care-giving. As expected, migrants having one additional child under the age of 15 increases both the in- whose spouses are inside tend to spend more time inside and circulate cident rate of return and being inside by about 8%. Having a spouse more often. Circularity is therefore a function of household arrange- inside increases the incident rate of return by 45% and being inside by ment: left-behind spouses and children encourage circular migration about 34%. Having a parent inside does not affect either dependent and longer time inside. While our data does not permit direct ob- variable significantly. servation of return reasons, our analyses confirm that circularity is re- In terms of rural housing, owning a brick-and-concrete house in- lated to individual and household factors. creases both the frequency of return and the time inside. This may be Our analyses also points to migrants’ increasing tendency to spend related to the migrants’ success in earning remittances that fund the longer periods outside. This may reflect the fact that new-generation houses and to the degree of satisfaction they have with the houses. A migrants have become the majority among rural-urban migrants and brick-and-concrete house tends to be the most desirable kind of house that they are more likely to migrate with a spouse than older migrants and a symbol of success, and in that light migrants may be motivated to (Liang et al., 2014; NPFPC, 2015; NBS, 2015). However, it is important return permanently, or at least more often and be inside longer. to note that the migration histories of rural Chinese are still evolving. We run the same models for the male sample in models (5) and (6), As migrants age, it is possible that return migration will increase and and for the female sample in models (7) and (8), in order to compare migrants will spend more time in the countryside (e.g. Jiang, 2013; how circularity varies by gender. Interestingly, being from the younger NPFPC, 2015). In addition, the dramatic increases in urban housing (mid and new) generations affects only a male migrant’s circularity but price in recent years have made it extremely difficult for migrant not a female migrant’s. However, when the first out-migration started workers to purchase a house in cities, especially in large cities. As long affects only a female migrant’s circularity but not a male migrant’s. as owning a good house continues to be an important goal for migrants Education level also affects only men but not women. These differences and their family members, circularity will persist. Absent pension and a to some extent suggest that rural men’s growing participation in the social safety net, circular migration is a reasonable practice of migrants urban labor force is sensitive to changes of their personal attributes, who can access urban work as well as maintain a satisfying living in the such as being new-generation or changes of their human capital, while countryside. rural women’s participation is more related to job opportunities in urban areas. Meanwhile, the number of children inside affects only Acknowledgements female migrants’ circularity by encouraging them to return more fre- quently and to spend more time inside, but has no effect on male mi- Dr. C. Cindy Fan would like to acknowledge the support by UCLA’s grants’ circularity. This finding underscores the persistence of gender Chancellor’sOffice and Division of Social Sciences. division of labor in Chinese rural households, namely, a wife is in charge of the inside work, such as taking care of children, while a Funding husband is in charge of the outside work, such as working in cities to provide financial support for the family (Jacka, 2006). This work was supported by the National Science Foundation [grant numbers BCS-0455107]. 8. Summary and conclusion References Different from the conventional focus on migration as primarily a step toward permanent settlement, this paper has foregrounded circu- Acevedo, G., Drachman, D., Paulino, A., 2004. Neither here nor there: Puerto Rican cir- larity as a core practice of rural Chinese, highlighting the frequency and cular migration. Immigrants Soc. Work Thinking Beyond Borders Uni. Bai, N., He, Y., 2003. Returning to the countryside versus continuing to work in the cities: duration of migrants’ returns. Using the inside-outside framework, a study on rural urban migrants and their return to the countryside of China. Soc. Sci. where inside refers to the rural and outside refers to the urban, our China 24, 149–158. fi Bastia, T., 2011. Should I stay or should I go? Return migration in times of crises. J. Int. analyses have also identi ed salient changes in rural-urban circularity Dev. 23, 583–595. over the past three decades. Drawing from the longitudinal migration Bedford, R., 1973. New Hebridean Mobility: A Study of Circular Migration. Australian histories of 530 rural-urban migrants from six villages in Anhui pro- National University. Bedford, R.D., 1980. Demographic processes in small islands: the case of internal mi- vince, and using two-level Poisson regression models, we have found gration. that in the more recent decade (2000–2009), migrants spent sig- Chan, C.K.-C., Ngai, P., Chan, J., 2010. The role of the state, labour policy and migrant ’ nificantly less amount of time inside (their home location). This sug- workers struggles in globalized China. In: Bowles, P., Harriss, J. (Eds.), Globalization and Labour in China and India: Impacts and Responses. Palgrave Macmillan UK, gests that urban job opportunities have become increasingly important London, pp. 45–63. to rural-urban migrants, and also that over time migrants have devel- Chan, K.W., 1996. Post-mao China: a two-class urban society in the making. Int. J. Urban Reg. Res. 20, 134–150. oped skills and strategies that allow them to stay longer in the city. Chan, K.W., 2010. The household registration system and migrant labor in China: notes Our analyses have also found that individual characteristics that on a debate. Popul. Develop. Rev. 36, 357–364. reduce the frequency of returns and the length of time inside include Cheng, D., 2008. Woguo jincheng nongmingong de juzhu wenti jiqi jiejue silu. [The Living Problem and Its Solutions for China's Migrant Workers] Zhongguo renkou, being male, new-generation, and not having attended senior high Ziyuan yu Huangjing (China Population, Rsources and Environment) 18, 78–84. school. Conversely, female migrants, older migrants and migrants with Clark, S.J., Collinson, M.A., Kahn, K., et al., 2007. Returning home to die: circular labour – senior high school education level tend to spend longer time inside. migration and mortality in South Africa 1. Scand. J. Public Health 35, 35 44.
103 C. Chen, C.C. Fan Geoforum 93 (2018) 97–104
Collinson, M., Tollman, S., Kahn, K., et al., 2006. Highly prevalent circular migration: 237–256. households, mobility and economic status in rural South Africa. Africa on the move: Massey, D.S., 1990. Social structure, household strategies, and the cumulative causation African migration and urbanisation in comparative perspective 194–216. of migration. Popul. Index 56, 3–26. Connelly, R., Roberts, K., Zheng, Z., 2010. The impact of circular migration on the po- Massey, D.S., Espinosa, K.E., 1997. What's driving Mexico-US migration? A theoretical, sition of married women in rural China. Feminist Econ. 16, 3–41. empirical, and policy analysis. Am. J. Sociol. 102, 939–999. Constant, A.F., Zimmermann, K.F., 2007. Circular migration: counts of exits and years Massey, D.S., Goldring, L., Durand, J., 1994. Continuities in transnational migration: an away from the host country. IZA Discussion Papers. analysis of nineteen Mexican communities. Am. J. Sociol. 1492–1533. Constant, A.F., Zimmermann, K.F., 2012. The dynamics of repeat migration: a Markov Massey, D.S., Pren, K.A., 2012. Unintended consequences of US immigration policy: ex- chain analysis. Int. Migr. Rev. 46, 362–388. plaining the post-1965 surge from Latin America. Popul. Develop. Rev. 38, 1–29. DaVanzo, J., 1983. Repeat migration in the United States: who moves back and who Mendoza, J.E., 2008. Economic and social determinants of Mexican circular and per- moves on? Rev. Econ. Stat. 65, 552–559. manent migration. Análisis Econ. 128, 124310–124323. Department of Land and Resources of Anhui Province, 2014. Anhui Sheng Dierci Tudi NBS, 2011. Xinshengdai Nongmingong de Shuliang, Jiegou he Tedian. Available at: Diaocha Zhuyao Shuju Gongbao.
104